Is the SaaSpocalypse nigh? The era of paying for software seats may be ending.

In December 2024, Microsoft CEO Satya Nadella appeared on the BG2 podcast and made a prediction that felt provocative and hyped. Business applications delivered as Software as a Service (SaaS) would collapse in the era of agentic AI, he said, because they are essentially CRUD databases layered on top of them. That logic would migrate to the AI tier, and the applications themselves would become irrelevant.

It sounded bold, and maybe even a bit dramatic. But almost a year later, we are witnessing that prediction playing out in real time.

Last week, Anthropic launched plugins for Cowork, its agentic desktop tool targeting knowledge workers. Within days, Thomson Reuters lost 16% of its stock value. RELX, the parent company of LexisNexis, dropped by 14%. Wolters Kluwer fell 10%. The selloff quickly spread beyond legal tech into Salesforce, ServiceNow, and Atlassian. Major Indian systems integrators like TCS and Infosys also saw a similar trend in their stock prices.

What’s intriguing about this is that a minor feature — a set of plugins launched by Anthropic — triggered the market sentiment. Financial analysts called this the SaaS apocalypse or “SaaSpocalypse.”

The thesis that became reality

A year ago, I wrote about the concept of “Service as Software” (SaS) for The New Stack. The framing, popularized by Foundation Capital, inverts the familiar and proven SaaS model. Traditional SaaS sells tools that enable humans to solve problems, whereas SaaS sells outcomes. AI agents do the actual work rather than just assisting on the sidelines.

For example, instead of QuickBooks providing forms and calculators to compute the payroll, an AI agent automatically runs the payroll each month without manual intervention. The same is true for tax, where an agent not only calculates taxes but also files returns using APIs and tools.

The selloff might not have been irrational panic. It very well could have been investors recognizing a structural shift.

At the time, this was just a thesis. A compelling one, but still mostly theoretical and forward-looking. Foundation Capital estimated the market opportunity at $4.6 trillion over five years, but the concrete products that deliver this capability were still emerging.

Anthropic’s Cowork plugins changed that.

What Cowork plugins actually are

Cowork launched in January as a simpler version of Claude Code for non-developers. It inherits Claude Code’s agentic architecture, which means Claude can take on complex, multi-step tasks, break them into sub-agents, coordinate parallel workstreams, and deliver finished outputs. The difference is that Cowork targets knowledge workers rather than engineers. Think document organization, research synthesis, and file management rather than code commits.

Plugins extend this foundation with domain expertise. Each plugin bundles skills, slash commands, and Model Context Protocol (MCP) connectors into a single installable package. The skills are markdown files that encode best practices, step-by-step workflows, and domain knowledge. The connectors integrate with enterprise systems like CRMs, document repositories, and databases. Slash commands give users quick access to everyday operations.

Anthropic launched 11 plugins covering productivity, sales, marketing, finance, legal, customer support, product management, enterprise search, data analysis, and biology research. There is also a meta-plugin for creating new plugins.

The legal plugin is what triggered the market panic. It automates contract review, NDA triage, compliance workflows, legal briefings, and templated responses. This is work that associates at law firms have billed at hundreds of dollars per hour for decades.

But here is the architectural insight that matters. Plugins are just markdown files. The domain expertise that took vertical SaaS companies years to accumulate and encode into complex software can now be expressed as skill configurations sitting on top of a general-purpose AI agent.

The Nadella connection

Let’s go back to what Nadella said in the podcast. He argued that SaaS applications are just CRUD databases with business logic, and agents would absorb that logic into the AI tier, making the application layer unnecessary. The backends would become interchangeable because agents would operate across multiple repositories without regard to which system they are updating.

Cowork plugins are literally that architecture made manifest.

The business logic of contract review is now a skill file. The integration with document management systems is an MCP connector. The workflows that legal software vendors spent years perfecting are now markdown instructions that Claude reads and executes. The application layer that Nadella predicted would collapse is precisely what the market priced in this week.

The selloff might not have been irrational panic. It very well could have been investors recognizing a structural shift.

Why this extends beyond legal

Legal tech took the biggest hit because Anthropic shipped a legal plugin, and the mapping to existing vendors is obvious. But the pattern applies everywhere.

The sales plugin helps with prospect research and deal preparation. The finance plugin analyzes financial data and builds forecasting models. The marketing plugin drafts content and plans campaigns. Each of these represents a category where specialized SaaS vendors have built businesses around encoding domain expertise into software.

What Cowork Plugins demonstrate is that a general-purpose AI provider can replicate those capabilities by packaging domain knowledge as configuration files. The moat that vertical SaaS companies built through years of product development, customer feedback, and workflow optimization has become much thinner.

IDC predicts that by 2028, pure seat-based pricing will be obsolete. According to CIO.com, 70% of software vendors will have to refocus their pricing on new value metrics such as consumption, outcomes, or organizational capability. The economics of enterprise software are shifting from selling access to tools toward selling delivered results.

The broader implications

The SaaSpocalypse is not a single event. It is the beginning of a structural transformation that will play out over the years.

For SaaS incumbents, the lesson is that domain expertise alone is no longer a defensible moat. If that expertise can be expressed as a set of prompts, skills, and workflow instructions, a foundation model provider or a well-funded competitor can replicate it quickly. The new moats will be built on data network effects, deep system integrations that agents cannot easily replicate, and outcome-based business models that align vendor incentives with customer results.

For enterprises, the opportunity is significant. Instead of paying per seat for tools that employees may or may not use effectively, they can pay for outcomes. The legal team does not need a contract review platform or training to use it. They need contracts reviewed. That shift in framing changes procurement, budgeting, and how organizations think about headcount.

For knowledge workers, the implications are more nuanced. The legal plugins explicitly note that licensed attorneys should review all outputs. These tools augment professionals rather than replacing them entirely, at least for now. But the volume of work a single professional can handle with AI assistance will increase dramatically, with downstream effects on staffing and career paths.

What’s next

Nadella made his prediction in December 2024. Thirteen months later, we have a concrete product delivered, along with a market reaction that validates the thesis.

The SaaSpocalypse is not about one company or one product. It is about the structural reality that business logic is migrating from application code to AI agents. Cowork plugins are the proof point, but they are not the endpoint.

The next few years will determine which SaaS vendors adapt and which become the next generation of legacy systems. The winners will be those who recognize that the value proposition has shifted from providing tools to delivering outcomes.

Service as Software is no longer a thesis. It is the evolving landscape.

The post Is the SaaSpocalypse nigh? The era of paying for software seats may be ending. appeared first on The New Stack.